Researchers at Berkeley Lab have developed GenomeOcean, an artificial intelligence model that reads DNA sequences the way ChatGPT reads text, uncovering hidden patterns and even predicting missing genetic code. This breakthrough represents a fundamentally new approach to genomics: instead of treating DNA as isolated data points, scientists are training AI to understand the "natural language" of genomes, opening doors to faster drug discovery, sustainable biofuels, and engineered biological systems.

What Is GenomeOcean and How Does It Work?

GenomeOcean is an AI-powered language model developed through a two-year collaboration between researchers at Berkeley Lab's Joint Genome Institute (JGI) and Northwestern University. Just as large language models (LLMs) like ChatGPT predict the next word in a sentence by analyzing patterns in billions of words, GenomeOcean learns from vast genomic sequences to identify patterns and relationships within DNA. The model was initially trained using NERSC's Perlmutter supercomputer, a high-performance system with powerful graphics processing units (GPUs) that enabled the computationally intensive pretraining phase.

What makes GenomeOcean particularly valuable is its ability to do more than just read genetic sequences. Like an autocomplete function on a smartphone, the model can fill in missing pieces of genetic code based on patterns it has learned from massive datasets. This capability is especially useful in synthetic biology, where scientists design new biological pathways for applications ranging from sustainable biofuels to pharmaceuticals and environmental solutions.

Why Should Scientists Care About AI That Understands Genetics?

The explosion of genomic data has created a bottleneck in biological research. Modern sequencing studies generate terabytes of data, far more than traditional computational tools can efficiently analyze. GenomeOcean addresses this challenge by leveraging AI to extract meaningful insights from these massive datasets. Researchers at the Joint BioEnergy Institute (JBEI), a Department of Energy Bioenergy Research Center managed by Berkeley Lab, are already considering GenomeOcean to improve the design of biological systems. By analyzing enormous datasets, the AI model can suggest new gene sequences that enhance productivity and efficiency in engineered biological pathways, potentially accelerating discoveries and reducing the trial-and-error cycle in laboratory experiments.

How to Leverage AI for Genomic Research: Key Infrastructure and Collaboration Steps

• Hybrid Computing Approach: Combine high-performance supercomputers like Perlmutter for the initial training phase with institutional GPU clusters for ongoing inference and predictions, balancing raw computational power with practical usability.
• Deep Cross-Functional Collaboration: Establish long-term partnerships between AI specialists and domain scientists through regular meetings and iterative refinement, rather than treating IT as a support function separate from research goals.
• Scalable Infrastructure Development: Build computational infrastructure specifically designed to support AI-powered genomic predictions, ensuring that models can be deployed and refined as research questions evolve.

The success of GenomeOcean hinges on a unique collaboration model between the JGI and Berkeley Lab's Science IT department. Rather than simply providing computational resources, Science IT has been embedded directly in the research process, participating in weekly meetings that sometimes run one to three hours and actively troubleshooting challenges alongside scientists.

"What makes this collaboration unique is not just the technology, but the long-term commitment. Science IT isn't just providing infrastructure; we're embedded in the research itself. If solving these large-scale AI challenges takes years, we'll be there every step of the way, working alongside scientists to refine, improve, and innovate," said Gary Jung, Head of the Science IT Department at Berkeley Lab.

Gary Jung, Head of the Science IT Department at Berkeley Lab

This partnership model highlights a broader shift in how scientific institutions approach AI development. Rather than treating AI as a tool to be deployed after the fact, successful genomic research requires AI specialists and biologists to work together from the beginning, refining proof-of-concept experiments and troubleshooting challenges in real time.

What Does This Mean for Medicine and Biotechnology?

The implications of GenomeOcean extend far beyond academic research. As more genomes are sequenced globally, the ability to catalog, analyze, and predict genomic functions will become essential for future discoveries in medicine, agriculture, and environmental science. In drug discovery, AI models that understand genetic patterns could accelerate the identification of therapeutic targets. In agriculture, they could help design crops with enhanced nutritional profiles or climate resilience. In environmental science, they could support the development of microorganisms engineered to break down pollutants or produce sustainable materials.

The work at Berkeley Lab demonstrates that combining AI, big data, and high-performance computing unlocks new possibilities in understanding life itself. GenomeOcean represents an exciting step toward a future where AI can help scientists read, write, and decode the language of life, making groundbreaking discoveries more accessible than ever before. Equally important is the hybrid collaboration model that makes such projects possible, where computing specialists play a long-term role not just in supporting infrastructure but in driving scientific breakthroughs.